High-grade serous ovarian cancer (HGSC) is the most common ovarian cancer subtype and the deadliest gynecological malignancy. Patients are diagnosed at a late stage, when the disease has already metastasized, and treated with surgery and platinum-taxane chemotherapy. Typically, after an initial complete response, the disease relapses and gradually becomes refractory to treatment. Individual patients, however, have extremely variable outcomes, which range from initial chemorefractiveness and short survival to complete cure.
It is of crucial importance to improve our understanding of the mechanisms underlying chemoresistance and to develop tests that help doctors to choose the optimal treatment for individual patients. However, outcome-associated biomarker identification is especially challenging in HGSC due to the complex genomic alterations characterizing this disease. The only common feature is inactivation of TP53, whereas oncogenic driver mutations are rare. In addition, homologous recombination (HR) DNA repair deficiency is detected in half of HGSC tumors; an HR defect indicates response to standard chemotherapy and novel PARP inhibitors (PARPi). About 20% of HGSCs carry CyclinE amplification, detectable by immunohistochemistry or in situ hybridization, which is linked to poor outcome.
As part of the ongoing EU-funded “Hercules” project, we are testing and validating tissue-based markers that may have prognostic implications. These markers have been identified by us and our collaborators. Among them is a panel identifying cancer stem cells that confer resistance to current chemotherapies. In these analyses, we use large HGSC cohorts, annotated pathology specimens and detailed longitudinal clinical information. The advanced biomarker analytics techniques include multiplex immunofluorescence imaging and RNAScope in situ hybridization.
Another approach for cancer prognostication includes novel image analysis technologies. As slide scanning technology has become better and faster, and the cost of data storage has decreased, novel utilization of digitized pathology slides (whole slide images) on a large scale has become feasible. Machine learning (ML), a branch of artificial intelligence (AI), uses computer algorithms to “see” scanned images of histologic tissue sections and learns to identify significant features.
This work, carried out in collaboration with Aiforia and Sampsa Hautaniemi’s group, intends to utilize a carefully constructed cohort of HGSC to evaluate whether machine learning can distinguish between groups of patients with very different clinical outcomes; a task not currently possible prospectively.
Our goals are to improve the clinical care of ovarian cancer patients by providing prognostic information, create a ML tool that can be used to address additional questions, and use AI/ML to go beyond replicating what a human pathologist can already do.
Germline DNA alterations predispose to a variety of diseases, including cancer. Identification of genetic disease associations requires analysis of large cohorts, which may be challenging in a prospective setting. Archived samples in pathology laboatories and biobanks, when combined with clinical information, represent a potential and so far neglected source for DNA array genotyping for genome-wide association studies (GWAS). In Finland, the FinnGen study aims to genotype blood samples from 500,000 individuals through Finnish biobanks. We are currently exploring, whether the cohort can be expanded with pathology specimens for selected disease entities.
Our project will compare genotyping results from DNA extracted from whole blood and formalin fixed paraffin embedded (FFPE) normal tissue specimens of the same patients, to optimize the DNA extraction methods, to evaluate the overall concordance, and to pilot the testing in patients with selected cancers. Successful implementation of the project could enhance identification of genetic factors contributing to disease risk on the scientific level, and optimize the budget allocated for disease screening and management on the societal level.
Despite the advances in early diagnosis and treatment achieved during the last ten years, the prognosis of metastatic colorectal cancer (mCRC) remains poor with a five-year relative survival rate of about 12%. Oncogenic signaling pathways, such as the EGFR pathway, and their regulators, including the cytoskeletal protein ezrin, are key determinants of cancer behavior and potential therapeutic targets. Better understanding of the biology of these pathways and mechanisms of resistance to targeted treatments is needed. We have previously shown that the response to anti-EGFR antibody treatment is critically dependent on, not just wild-type RAS pathway, but also EGFR copy number gain. Our aim is to better understand, why this, typically heterogenous, EGFR amplification facilitates tumor growth.
The objectives of this project are first, to identify the downstream pro-oncogenic EGFR signaling pathways in patients with RAS/BRAF/PIK3CA wild type mCRC, that show a heterogeneous distribution of the EGFR Gene Copy Number (GCN) and elucidate the relationship between different tumor cell subsets, in an attempt to explain the mechanisms defining anti-EGFR therapy response in RAS/BRAF/PIK3CA wild type mCRC.
Another objective of this project is to study the co-expression of Ezrin and Podocalyxin, molecules whose upregulation has been linked with EMT (epithelial to mesenchymal transition) and to determine, whether EGFR signaling is related to Ezrin and Podocalyxin upregulation.
This biobank research project, carried out in collaboration with Auria biobank and oncologists at Turku University Hospital, aims to apply open source image processing algorithms for prognostication of stage II colorectal cancer patients. We will evaluate, AI tools could help pathologists and clinicians to identify patients at high-risk of disease relapse. In the future, chemotherapy could then be targeted to the right patients.